Simulated and Real Robot Interaction Data

该数据集包含了模拟的机器人操作任务视频，这些任务涉及刚体和非刚体对象的操作，其目的是为了评估通过对比域随机化学习到的表征。该数据集使用了Bullet物理模拟器生成的模拟图像，并从可描述纹理数据集（DTD）中随机采样纹理进行渲染。该数据集包含了4,000个视频，每个视频有15帧，用于在机器人任务中进行模拟到真实的评估，以推进机器人操作和表征学习的研究。

This dataset comprises simulated robotic manipulation task videos, where the tasks involve manipulating both rigid and non-rigid objects. Its core objective is to evaluate representations learned via contrastive domain randomization. The dataset uses simulated images generated by the Bullet Physics Simulator, and randomly samples textures from the Describable Textures Dataset (DTD) for rendering. This dataset includes 4,000 videos, each with 15 frames, and is intended for sim-to-real evaluation in robotic tasks to advance research in robotic manipulation and representation learning.